Add PMEM support for RISC-V
v2 cover Add PMEM support for RISC-V
The Linux NVDIMM PEM drivers require arch support to map and access the persistent memory device. This series adds RISC-V PMEM support using recently added Svpbmt and Zicbom support.
These patches can also be found in riscv_pmem_v2 branch at:
https://github.com/avpatel/linux.git
Changes since v1:
- Fix error reported by test bot
https://lore.kernel.org/all/[email protected]/
Anup Patel (4):
RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
RISC-V: Move riscv_init_cbom_blocksize() to cacheflush.c
RISC-V: Implement arch specific PMEM APIs
RISC-V: Enable PMEM drivers
arch/riscv/Kconfig | 1 + arch/riscv/configs/defconfig | 1 + arch/riscv/include/asm/cacheflush.h | 2 ++ arch/riscv/include/asm/io.h | 10 ++++++++ arch/riscv/include/asm/pgtable.h | 2 ++ arch/riscv/mm/Makefile | 1 + arch/riscv/mm/cacheflush.c | 39 +++++++++++++++++++++++++++++ arch/riscv/mm/dma-noncoherent.c | 38 ---------------------------- arch/riscv/mm/pmem.c | 21 ++++++++++++++++ 9 files changed, 77 insertions(+), 38 deletions(-) create mode 100644 arch/riscv/mm/pmem.c
Signature · 1 line
2.34.1
PATCH v2 1/4 RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
Currently, all flavors of ioremap_xyz() function maps to the generic ioremap() which means any ioremap_xyz() call will always map the target memory as IO using _PAGE_IOREMAP page attributes. This breaks ioremap_cache() and ioremap_wc() on systems with Svpbmt because memory remapped using ioremap_cache() and ioremap_wc() will use _PAGE_IOREMAP page attributes.
To address above (just like other architectures), we implement RISC-V specific ioremap_cache() and ioremap_wc() which maps memory using page attributes as defined by the Svpbmt specification.
--- arch/riscv/include/asm/io.h | 10 ++++++++++ arch/riscv/include/asm/pgtable.h | 2 ++ 2 files changed, 12 insertions(+)
arch/riscv/include/asm/io.h +10 -0
@@ -133,6 +133,16 @@ __io_writes_outs(outs, u64, q, __io_pbr(), __io_paw())#define outsq (addr, buffer, count) __outsq((void __iomem *)addr, buffer, count)#endif +#ifdef CONFIG_MMU +#define ioremap_wc (addr, size) \ + ioremap_prot((addr), (size), _PAGE_IOREMAP_WC) +#endif +#include <asm-generic/io.h> +#ifdef CONFIG_MMU +#define ioremap_cache (addr, size) \ + ioremap_prot((addr), (size), _PAGE_KERNEL) +#endif +#endif /* _ASM_RISCV_IO_H */
arch/riscv/include/asm/pgtable.h +2 -0
@@ -182,6 +182,8 @@ extern struct pt_alloc_ops pt_ops __initdata;#define PAGE_TABLE __pgprot(_PAGE_TABLE)#define _PAGE_IOREMAP ((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO) +#define _PAGE_IOREMAP_WC ((_PAGE_KERNEL & ~_PAGE_MTMASK) | \ + _PAGE_NOCACHE)#define PAGE_KERNEL_IO __pgprot(_PAGE_IOREMAP)extern pgd_t swapper_pg_dir[];
Signature · 1 line
2.34.1
Re: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
Am Dienstag, 30. August 2022, 06:46:39 CEST schrieb Anup Patel:
Quoted 7 lines
Currently, all flavors of
ioremap_xyz()function maps to the genericioremap()which means anyioremap_xyz()call will always map the target memory as IO using_PAGE_IOREMAPpage attributes. This breaksioremap_cache()andioremap_wc()on systems with Svpbmt because memory remapped usingioremap_cache()andioremap_wc()will use_PAGE_IOREMAPpage attributes.
Quoted 8 lines
To address above (just like other architectures), we implement RISC-V specific
ioremap_cache()andioremap_wc()which maps memory using page attributes as defined by the Svpbmt specification.Fixes ff689fd21cb1 ("riscv: add RISC-V Svpbmt extension support")Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>
And at least on a qemu
But it was running there before as well of course.
Heiko
Quoted 32 lines
--- arch/riscv/include/asm/io.h | 10 ++++++++++ arch/riscv/include/asm/pgtable.h | 2 ++ 2 files changed, 12 insertions(+)arch/riscv/include/asm/io.h +10
-0@@ -133,6 +133,16 @@ __io_writes_outs(outs, u64, q, __io_pbr(), __io_paw())#define outsq (addr, buffer, count) __outsq((void __iomem *)addr, buffer, count)#endif +#ifdef CONFIG_MMU +#define ioremap_wc (addr, size) \ + ioremap_prot((addr), (size), _PAGE_IOREMAP_WC) +#endif +#include <asm-generic/io.h> +#ifdef CONFIG_MMU +#define ioremap_cache (addr, size) \ + ioremap_prot((addr), (size), _PAGE_KERNEL) +#endif +#endif /* _ASM_RISCV_IO_H */ arch/riscv/include/asm/pgtable.h +0
-0
arch/riscv/include/asm/pgtable.h extern struct pt_alloc_ops pt_ops __initdata; · line 189
Quoted 8 lines
arch/riscv/include/asm/pgtable.h +2
-0@@ -182,6 +182,8 @@ extern struct pt_alloc_ops pt_ops __initdata;#define PAGE_TABLE __pgprot(_PAGE_TABLE)#define _PAGE_IOREMAP ((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO) +#define _PAGE_IOREMAP_WC ((_PAGE_KERNEL & ~_PAGE_MTMASK) | \ + _PAGE_NOCACHE)#define PAGE_KERNEL_IO __pgprot(_PAGE_IOREMAP)extern pgd_t swapper_pg_dir[];
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Re: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
In reply to Anup PatelQuoted 7 lines · Anup Patel
Currently, all flavors of
ioremap_xyz()function maps to the genericioremap()which means anyioremap_xyz()call will always map the target memory as IO using_PAGE_IOREMAPpage attributes. This breaksioremap_cache()andioremap_wc()on systems with Svpbmt because memory remapped usingioremap_cache()andioremap_wc()will use_PAGE_IOREMAPpage attributes.
Quoted 8 lines
To address above (just like other architectures), we implement RISC-V specific
ioremap_cache()andioremap_wc()which maps memory using page attributes as defined by the Svpbmt specification.Fixes ff689fd21cb1 ("riscv: add RISC-V Svpbmt extension support")Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>
Seems sane to me too :)
Quoted 39 lines
--- arch/riscv/include/asm/io.h | 10 ++++++++++ arch/riscv/include/asm/pgtable.h | 2 ++ 2 files changed, 12 insertions(+)arch/riscv/include/asm/io.h +10
-0@@ -133,6 +133,16 @@ __io_writes_outs(outs, u64, q, __io_pbr(), __io_paw())#define outsq (addr, buffer, count) __outsq((void __iomem *)addr, buffer, count)#endif +#ifdef CONFIG_MMU +#define ioremap_wc (addr, size) \ + ioremap_prot((addr), (size), _PAGE_IOREMAP_WC) +#endif +#include <asm-generic/io.h> +#ifdef CONFIG_MMU +#define ioremap_cache (addr, size) \ + ioremap_prot((addr), (size), _PAGE_KERNEL) +#endif +#endif /* _ASM_RISCV_IO_H */ arch/riscv/include/asm/pgtable.h +2
-0@@ -182,6 +182,8 @@ extern struct pt_alloc_ops pt_ops __initdata;#define PAGE_TABLE __pgprot(_PAGE_TABLE)#define _PAGE_IOREMAP ((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO) +#define _PAGE_IOREMAP_WC ((_PAGE_KERNEL & ~_PAGE_MTMASK) | \ + _PAGE_NOCACHE)#define PAGE_KERNEL_IO __pgprot(_PAGE_IOREMAP)extern pgd_t swapper_pg_dir[];
Re: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
In reply to Anup PatelHi Palmer,
Quoted 8 lines · Anup Patel
Currently, all flavors of
ioremap_xyz()function maps to the genericioremap()which means anyioremap_xyz()call will always map the target memory as IO using_PAGE_IOREMAPpage attributes. This breaksioremap_cache()andioremap_wc()on systems with Svpbmt because memory remapped usingioremap_cache()andioremap_wc()will use_PAGE_IOREMAPpage attributes.
Quoted 8 lines
To address above (just like other architectures), we implement RISC-V specific
ioremap_cache()andioremap_wc()which maps memory using page attributes as defined by the Svpbmt specification.Fixes ff689fd21cb1 ("riscv: add RISC-V Svpbmt extension support")Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>
Can you please consider this patch for Linux-6.0-rc5 ?
Regards,
Anup
Quoted 42 lines
--- arch/riscv/include/asm/io.h | 10 ++++++++++ arch/riscv/include/asm/pgtable.h | 2 ++ 2 files changed, 12 insertions(+)arch/riscv/include/asm/io.h +10
-0@@ -133,6 +133,16 @@ __io_writes_outs(outs, u64, q, __io_pbr(), __io_paw())#define outsq (addr, buffer, count) __outsq((void __iomem *)addr, buffer, count)#endif +#ifdef CONFIG_MMU +#define ioremap_wc (addr, size) \ + ioremap_prot((addr), (size), _PAGE_IOREMAP_WC) +#endif +#include <asm-generic/io.h> +#ifdef CONFIG_MMU +#define ioremap_cache (addr, size) \ + ioremap_prot((addr), (size), _PAGE_KERNEL) +#endif +#endif /* _ASM_RISCV_IO_H */ arch/riscv/include/asm/pgtable.h +2
-1@@ -182,6 +182,8 @@ extern struct pt_alloc_ops pt_ops __initdata;#define PAGE_TABLE __pgprot(_PAGE_TABLE)#define _PAGE_IOREMAP ((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO) +#define _PAGE_IOREMAP_WC ((_PAGE_KERNEL & ~_PAGE_MTMASK) | \ + _PAGE_NOCACHE)#define PAGE_KERNEL_IO __pgprot(_PAGE_IOREMAP)extern pgd_t swapper_pg_dir[]; --2.34.1
Re: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
In reply to Anup PatelHi Palmer,
Quoted 8 lines · Anup Patel
Currently, all flavors of
ioremap_xyz()function maps to the genericioremap()which means anyioremap_xyz()call will always map the target memory as IO using_PAGE_IOREMAPpage attributes. This breaksioremap_cache()andioremap_wc()on systems with Svpbmt because memory remapped usingioremap_cache()andioremap_wc()will use_PAGE_IOREMAPpage attributes.
Quoted 8 lines
To address above (just like other architectures), we implement RISC-V specific
ioremap_cache()andioremap_wc()which maps memory using page attributes as defined by the Svpbmt specification.Fixes ff689fd21cb1 ("riscv: add RISC-V Svpbmt extension support")Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>
This is a crucial RC fix. Can you please take this ?
Regards,
Anup
Quoted 42 lines
--- arch/riscv/include/asm/io.h | 10 ++++++++++ arch/riscv/include/asm/pgtable.h | 2 ++ 2 files changed, 12 insertions(+)arch/riscv/include/asm/io.h +10
-0@@ -133,6 +133,16 @@ __io_writes_outs(outs, u64, q, __io_pbr(), __io_paw())#define outsq (addr, buffer, count) __outsq((void __iomem *)addr, buffer, count)#endif +#ifdef CONFIG_MMU +#define ioremap_wc (addr, size) \ + ioremap_prot((addr), (size), _PAGE_IOREMAP_WC) +#endif +#include <asm-generic/io.h> +#ifdef CONFIG_MMU +#define ioremap_cache (addr, size) \ + ioremap_prot((addr), (size), _PAGE_KERNEL) +#endif +#endif /* _ASM_RISCV_IO_H */ arch/riscv/include/asm/pgtable.h +2
-1@@ -182,6 +182,8 @@ extern struct pt_alloc_ops pt_ops __initdata;#define PAGE_TABLE __pgprot(_PAGE_TABLE)#define _PAGE_IOREMAP ((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO) +#define _PAGE_IOREMAP_WC ((_PAGE_KERNEL & ~_PAGE_MTMASK) | \ + _PAGE_NOCACHE)#define PAGE_KERNEL_IO __pgprot(_PAGE_IOREMAP)extern pgd_t swapper_pg_dir[]; --2.34.1
Palmer Dabbelt
maintainerRe: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
Quoted 15 lines · On Thu, 15 Sep 2022 19:24:55 PDT (-0700), [email protected] wrote:
Hi Palmer,
Quoted 12 lines · Anup Patel
Currently, all flavors of
ioremap_xyz()function maps to the genericioremap()which means anyioremap_xyz()call will always map the target memory as IO using_PAGE_IOREMAPpage attributes. This breaksioremap_cache()andioremap_wc()on systems with Svpbmt because memory remapped usingioremap_cache()andioremap_wc()will use_PAGE_IOREMAPpage attributes.To address above (just like other architectures), we implement RISC-V specific
ioremap_cache()andioremap_wc()which maps memory using page attributes as defined by the Svpbmt specification.
Quoted 6 lines
Quoted 4 lines
Fixes ff689fd21cb1 ("riscv: add RISC-V Svpbmt extension support")Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>This is a crucial RC fix. Can you please take this ?
Sorry I missed this, I thought it was just part of the rest of this patch set. That said, I'm not actually sure this is a critical fix: sure it's a performance problem, and if some driver is expecting ioremap_cache() to go fast then possibly a pretty big one, but the only Svpmbt hardware that exists is the D1 and that was just supported this release so it's not a regression. Maybe that's a bit pedantic, but all this travel has kind of made things a mess and I'm trying to make sure nothing goes off the rails.
Probably a pointless distinction as it'll just get backported anyway, but I'm going to hold off on this for now -- it generally looks OK, but I don't get back until this weekend and I'm super tired so I'm trying to avoid screwing anything up.
Quoted 46 lines
Regards,
AnupQuoted 42 lines
--- arch/riscv/include/asm/io.h | 10 ++++++++++ arch/riscv/include/asm/pgtable.h | 2 ++ 2 files changed, 12 insertions(+)arch/riscv/include/asm/io.h +10
-0@@ -133,6 +133,16 @@ __io_writes_outs(outs, u64, q, __io_pbr(), __io_paw())#define outsq (addr, buffer, count) __outsq((void __iomem *)addr, buffer, count)#endif +#ifdef CONFIG_MMU +#define ioremap_wc (addr, size) \ + ioremap_prot((addr), (size), _PAGE_IOREMAP_WC) +#endif +#include <asm-generic/io.h> +#ifdef CONFIG_MMU +#define ioremap_cache (addr, size) \ + ioremap_prot((addr), (size), _PAGE_KERNEL) +#endif +#endif /* _ASM_RISCV_IO_H */ arch/riscv/include/asm/pgtable.h +2
-1@@ -182,6 +182,8 @@ extern struct pt_alloc_ops pt_ops __initdata;#define PAGE_TABLE __pgprot(_PAGE_TABLE)#define _PAGE_IOREMAP ((_PAGE_KERNEL & ~_PAGE_MTMASK) | _PAGE_IO) +#define _PAGE_IOREMAP_WC ((_PAGE_KERNEL & ~_PAGE_MTMASK) | \ + _PAGE_NOCACHE)#define PAGE_KERNEL_IO __pgprot(_PAGE_IOREMAP)extern pgd_t swapper_pg_dir[]; --2.34.1
Re: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
Quoted 22 lines · Palmer Dabbelt
Quoted 14 lines · On Thu, 15 Sep 2022 19:24:55 PDT (-0700), [email protected] wrote:
Quoted 12 lines · Anup Patel
Currently, all flavors of
ioremap_xyz()function maps to the genericioremap()which means anyioremap_xyz()call will always map the target memory as IO using_PAGE_IOREMAPpage attributes. This breaksioremap_cache()andioremap_wc()on systems with Svpbmt because memory remapped usingioremap_cache()andioremap_wc()will use_PAGE_IOREMAPpage attributes.To address above (just like other architectures), we implement RISC-V specific
ioremap_cache()andioremap_wc()which maps memory using page attributes as defined by the Svpbmt specification.Quoted 6 lines
Quoted 4 lines
Fixes ff689fd21cb1 ("riscv: add RISC-V Svpbmt extension support")Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>This is a crucial RC fix. Can you please take this ?
Quoted 8 lines
Sorry I missed this, I thought it was just part of the rest of this patch set. That said, I'm not actually sure this is a critical fix:
sure it's a performance problem, and if some driver is expectingioremap_cache()to go fast then possibly a pretty big one, but the only Svpmbt hardware that exists is the D1 and that was just supported this release so it's not a regression. Maybe that's a bit pedantic, but all this travel has kind of made things a mess and I'm trying to make sure nothing goes off the rails.
I think generally speaking any use of ioremap_cache() in a driver is a mistake. The few users that exist are usually from historic x86 specific code and are hard to kill off.
Arnd
Palmer Dabbelt
maintainerRe: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
Quoted 32 lines · Arnd Bergmann
Quoted 22 lines · Palmer Dabbelt
Quoted 14 lines · On Thu, 15 Sep 2022 19:24:55 PDT (-0700), [email protected] wrote:
Quoted 12 lines · Anup Patel
Currently, all flavors of
ioremap_xyz()function maps to the genericioremap()which means anyioremap_xyz()call will always map the target memory as IO using_PAGE_IOREMAPpage attributes. This breaksioremap_cache()andioremap_wc()on systems with Svpbmt because memory remapped usingioremap_cache()andioremap_wc()will use_PAGE_IOREMAPpage attributes.To address above (just like other architectures), we implement RISC-V specific
ioremap_cache()andioremap_wc()which maps memory using page attributes as defined by the Svpbmt specification.Quoted 6 lines
Quoted 4 lines
Fixes ff689fd21cb1 ("riscv: add RISC-V Svpbmt extension support")Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>This is a crucial RC fix. Can you please take this ?
Quoted 8 lines
Sorry I missed this, I thought it was just part of the rest of this patch set. That said, I'm not actually sure this is a critical fix:
sure it's a performance problem, and if some driver is expectingioremap_cache()to go fast then possibly a pretty big one, but the only Svpmbt hardware that exists is the D1 and that was just supported this release so it's not a regression. Maybe that's a bit pedantic, but all this travel has kind of made things a mess and I'm trying to make sure nothing goes off the rails.
Quoted 3 lines
I think generally speaking any use of
ioremap_cache()in a driver is a mistake. The few users that exist are usually from historic x86 specific code and are hard to kill off.
Should we just add some sort of CONFIG_ARCH_HAS_IOREMAP_CACHE and then ban those drivers from everywhere else?
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Re: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
In reply to Palmer DabbeltQuoted 4 lines · Palmer Dabbelt
Sorry I missed this, I thought it was just part of the rest of this patch set. That said, I'm not actually sure this is a critical fix: sure it's a performance problem, and if some driver is expecting
ioremap_cache()to go fast then possibly a pretty big one, but the only Svpmbt hardware that
More importantly nothing should be using ioremap_cache at all. It is an API that has long been deprecated in favor of memremap.
Palmer Dabbelt
maintainerRe: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
Quoted 8 lines · Christoph Hellwig
Quoted 4 lines · Palmer Dabbelt
Sorry I missed this, I thought it was just part of the rest of this patch set. That said, I'm not actually sure this is a critical fix: sure it's a performance problem, and if some driver is expecting
ioremap_cache()to go fast then possibly a pretty big one, but the only Svpmbt hardware thatMore importantly nothing should be using
ioremap_cacheat all. It is an API that has long been deprecated in favor of memremap.
There's a few uses of it, I just send along a patch to make it arch-specific and make the users depend on it. That should let us stop adding this to ports just to get the drivers building.
Re: [PATCH v2 1/4] RISC-V: Fix ioremap_cache() and ioremap_wc() for systems with Svpbmt
Quoted 8 lines · Palmer Dabbelt
Quoted 6 lines · Christoph Hellwig
Quoted 4 lines · Palmer Dabbelt
Sorry I missed this, I thought it was just part of the rest of this patch set. That said, I'm not actually sure this is a critical fix: sure it's a performance problem, and if some driver is expecting
ioremap_cache()to go fast then possibly a pretty big one, but the only Svpmbt hardware that
Quoted 6 lines
Quoted 2 lines
More importantly nothing should be using
ioremap_cacheat all. It is an API that has long been deprecated in favor of memremap.There's a few uses of it, I just send along a patch to make it arch-specific and make the users depend on it. That should let us stop adding this to ports just to get the drivers building.
I agree, ioremap_cache() should not be used by drivers.
I encountered this issue with the PMEM driver which uses devm_memremap() which in-turn calls memremap() (kernel/iomem.c). The kernel memremap() still expects arch-specific ioremap_xyz() functions/macros to implement various MEMREMAP_xyz flags which is why we need these RISC-V specific ioremap_xyz().
An alternative way is to implement RISC-V specific arch_memremap_wb() but this will still look similar to ioremap_cache() added by this patch. Also, only 32bit ARM implements arch_memremap_wb() whereas all other archs (ARM64, x86_64, etc) implement ioremap_xyz() functions/macros.
Regards,
Anup
PATCH v2 2/4 RISC-V: Move riscv_init_cbom_blocksize() to cacheflush.c
In reply to Anup PatelThe riscv_cbom_block_size parsing from DT belongs to cacheflush.c which is home for all cache maintenance related stuff so let us move the riscv_init_cbom_blocksize() and riscv_cbom_block_size to cacheflush.c.
--- arch/riscv/include/asm/cacheflush.h | 2 ++ arch/riscv/mm/cacheflush.c | 39 +++++++++++++++++++++++++++++ arch/riscv/mm/dma-noncoherent.c | 38 ---------------------------- 3 files changed, 41 insertions(+), 38 deletions(-)
arch/riscv/include/asm/cacheflush.h +2 -0
@@ -42,6 +42,8 @@ void flush_icache_mm(struct mm_struct *mm, bool local);#endif /* CONFIG_SMP */ +extern unsigned int riscv_cbom_block_size; +#ifdef CONFIG_RISCV_ISA_ZICBOM void riscv_init_cbom_blocksize (void );#else
arch/riscv/mm/cacheflush.c +39 -0
@@ -3,6 +3,8 @@* Copyright (C) 2017 SiFive */ +#include <linux/of.h> +#include <linux/of_device.h> #include <asm/cacheflush.h> #ifdef CONFIG_SMP
@@ -86,3 +88,40 @@ void flush_icache_pte(pte_t pte) flush_icache_all(); }#endif /* CONFIG_MMU */ + +unsigned int riscv_cbom_block_size =L1_CACHE_BYTES ; + +#ifdef CONFIG_RISCV_ISA_ZICBOM +void riscv_init_cbom_blocksize (void ) +{ +struct device_node *node; +int ret; +u32 val; + +for_each_of_cpu_node (node) { +unsigned long hartid; +int cbom_hartid; + + ret =riscv_of_processor_hartid (node, &hartid); +if (ret) +continue ; + +if (hartid <0 ) +continue ; + +/* set block-size for cbom extension if available */ + ret =of_property_read_u32 (node,"riscv,cbom-block-size" , &val); +if (ret) +continue ; + +if (!riscv_cbom_block_size) { + riscv_cbom_block_size = val; + cbom_hartid = hartid; + }else { +if (riscv_cbom_block_size != val) +pr_warn ("cbom-block-size mismatched between harts %d and %lu\n" , + cbom_hartid, hartid); + } + } +} +#endif
arch/riscv/mm/dma-noncoherent.c +0 -38
@@ -8,11 +8,8 @@#include <linux/dma-direct.h> #include <linux/dma-map-ops.h> #include <linux/mm.h> -#include <linux/of.h> -#include <linux/of_device.h> #include <asm/cacheflush.h> -static unsigned int riscv_cbom_block_size =L1_CACHE_BYTES ;static bool noncoherent_supported; void arch_sync_dma_for_device(phys_addr_t paddr, size_t size,
@@ -75,41 +72,6 @@ void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size, dev->dma_coherent = coherent; } -#ifdef CONFIG_RISCV_ISA_ZICBOM -void riscv_init_cbom_blocksize (void ) -{ -struct device_node *node; -int ret; -u32 val; - -for_each_of_cpu_node (node) { -unsigned long hartid; -int cbom_hartid; - - ret =riscv_of_processor_hartid (node, &hartid); -if (ret) -continue ; - -if (hartid <0 ) -continue ; - -/* set block-size for cbom extension if available */ - ret =of_property_read_u32 (node,"riscv,cbom-block-size" , &val); -if (ret) -continue ; - -if (!riscv_cbom_block_size) { - riscv_cbom_block_size = val; - cbom_hartid = hartid; - }else { -if (riscv_cbom_block_size != val) -pr_warn ("cbom-block-size mismatched between harts %d and %lu\n" , - cbom_hartid, hartid); - } - } -} -#endif -void riscv_noncoherent_supported (void ) { noncoherent_supported = true;
Signature · 1 line
2.34.1
Re: [PATCH v2 2/4] RISC-V: Move riscv_init_cbom_blocksize() to cacheflush.c
Hi,
Am Dienstag, 30. August 2022, 06:46:40 CEST schrieb Anup Patel:
Quoted 7 lines
The
riscv_cbom_block_sizeparsing from DT belongs tocacheflush.cwhich is home for all cache maintenance related stuff so let us move theriscv_init_cbom_blocksize()andriscv_cbom_block_sizetocacheflush.c.Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>
Makes a lot of sense to keep stuff together.
Also, can we handle this as fix patch?
I.e. Currently the t-head code somewhat relies on the default value set to L1_CACHE_BYTES. The cache-block-size is static there.
Palmers upcoming patch reworking the parsing [0], will remove that default, so having the riscv_cbom_block_size defined in the cacheflush header will allow an easy fix by setting that value from the t-head errata init for those cores.
Heiko
[0] https://lore.kernel.org/r/[email protected]
Quoted 122 lines
--- arch/riscv/include/asm/cacheflush.h | 2 ++ arch/riscv/mm/cacheflush.c | 39 +++++++++++++++++++++++++++++ arch/riscv/mm/dma-noncoherent.c | 38 ---------------------------- 3 files changed, 41 insertions(+), 38 deletions(-)arch/riscv/include/asm/cacheflush.h +2
-0@@ -42,6 +42,8 @@ void flush_icache_mm(struct mm_struct *mm, bool local);#endif /* CONFIG_SMP */ +extern unsigned int riscv_cbom_block_size; +#ifdef CONFIG_RISCV_ISA_ZICBOM void riscv_init_cbom_blocksize (void );#else arch/riscv/mm/cacheflush.c +39
-0@@ -3,6 +3,8 @@* Copyright (C) 2017 SiFive */ +#include <linux/of.h> +#include <linux/of_device.h> #include <asm/cacheflush.h> #ifdef CONFIG_SMP @@ -86,3 +88,40 @@ void flush_icache_pte(pte_t pte) flush_icache_all(); }#endif /* CONFIG_MMU */ + +unsigned int riscv_cbom_block_size =L1_CACHE_BYTES ; + +#ifdef CONFIG_RISCV_ISA_ZICBOM +void riscv_init_cbom_blocksize (void ) +{ +struct device_node *node; +int ret; +u32 val; + +for_each_of_cpu_node (node) { +unsigned long hartid; +int cbom_hartid; + + ret =riscv_of_processor_hartid (node, &hartid); +if (ret) +continue ; + +if (hartid <0 ) +continue ; + +/* set block-size for cbom extension if available */ + ret =of_property_read_u32 (node,"riscv,cbom-block-size" , &val); +if (ret) +continue ; + +if (!riscv_cbom_block_size) { + riscv_cbom_block_size = val; + cbom_hartid = hartid; + }else { +if (riscv_cbom_block_size != val) +pr_warn ("cbom-block-size mismatched between harts %d and %lu\n" , + cbom_hartid, hartid); + } + } +} +#endif arch/riscv/mm/dma-noncoherent.c +0
-31@@ -8,11 +8,8 @@#include <linux/dma-direct.h> #include <linux/dma-map-ops.h> #include <linux/mm.h> -#include <linux/of.h> -#include <linux/of_device.h> #include <asm/cacheflush.h> -static unsigned int riscv_cbom_block_size =L1_CACHE_BYTES ;static bool noncoherent_supported; void arch_sync_dma_for_device(phys_addr_t paddr, size_t size,@@ -75,41 +72,6 @@ void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size, dev->dma_coherent = coherent; } -#ifdef CONFIG_RISCV_ISA_ZICBOM -void riscv_init_cbom_blocksize (void ) -{ -struct device_node *node; -int ret; -u32 val; - -for_each_of_cpu_node (node) { -unsigned long hartid; -int cbom_hartid; - - ret =riscv_of_processor_hartid (node, &hartid); -if (ret) -continue ; - -if (hartid <0 ) -continue ; - -/* set block-size for cbom extension if available */ - ret =of_property_read_u32 (node,"riscv,cbom-block-size" , &val); -if (ret) -continue ; - -if (!riscv_cbom_block_size) { - riscv_cbom_block_size = val; - cbom_hartid = hartid; - }else { -if (riscv_cbom_block_size != val)
arch/riscv/mm/dma-noncoherent.c void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size, · line 77
Quoted 10 lines
arch/riscv/mm/dma-noncoherent.c +0
-7-pr_warn ("cbom-block-size mismatched between harts %d and %lu\n" , - cbom_hartid, hartid); - } - } -} -#endif -void riscv_noncoherent_supported (void ) { noncoherent_supported = true;
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
[email protected]
Reviewed-byRe: [PATCH v2 2/4] RISC-V: Move riscv_init_cbom_blocksize() to cacheflush.c
Quoted 19 lines · Heiko Stübner
Hi,
Am Dienstag, 30. August 2022, 06:46:40 CEST schrieb Anup Patel:
Quoted 7 lines
The
riscv_cbom_block_sizeparsing from DT belongs tocacheflush.cwhich is home for all cache maintenance related stuff so let us move theriscv_init_cbom_blocksize()andriscv_cbom_block_sizetocacheflush.c.Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>Makes a lot of sense to keep stuff together.
Reviewed-by Heiko Stuebner <[email protected]>Tested-by Heiko Stuebner <[email protected]>Also, can we handle this as fix patch?
Quoted 7 lines
I.e. Currently the t-head code somewhat relies on the default value set to
L1_CACHE_BYTES. The cache-block-size is static there.Palmers upcoming patch reworking the parsing [0], will remove that default, so having the
riscv_cbom_block_sizedefined in the cacheflush header will allow an easy fix by setting that value from the t-head errata init for those cores.
@Palmer what is the status of that fix?
Since I have a clang toolchain set up, I could squash my fixup-for-clang into your patch and respin it as a real patch for Anup to base on?
Anyway, I too like this cleanup:
Quoted 147 lines
Heiko
[0] https://lore.kernel.org/r/[email protected]
Quoted 122 lines
--- arch/riscv/include/asm/cacheflush.h | 2 ++ arch/riscv/mm/cacheflush.c | 39 +++++++++++++++++++++++++++++ arch/riscv/mm/dma-noncoherent.c | 38 ---------------------------- 3 files changed, 41 insertions(+), 38 deletions(-)arch/riscv/include/asm/cacheflush.h +2
-0@@ -42,6 +42,8 @@ void flush_icache_mm(struct mm_struct *mm, bool local);#endif /* CONFIG_SMP */ +extern unsigned int riscv_cbom_block_size; +#ifdef CONFIG_RISCV_ISA_ZICBOM void riscv_init_cbom_blocksize (void );#else arch/riscv/mm/cacheflush.c +39
-0@@ -3,6 +3,8 @@* Copyright (C) 2017 SiFive */ +#include <linux/of.h> +#include <linux/of_device.h> #include <asm/cacheflush.h> #ifdef CONFIG_SMP @@ -86,3 +88,40 @@ void flush_icache_pte(pte_t pte) flush_icache_all(); }#endif /* CONFIG_MMU */ + +unsigned int riscv_cbom_block_size =L1_CACHE_BYTES ; + +#ifdef CONFIG_RISCV_ISA_ZICBOM +void riscv_init_cbom_blocksize (void ) +{ +struct device_node *node; +int ret; +u32 val; + +for_each_of_cpu_node (node) { +unsigned long hartid; +int cbom_hartid; + + ret =riscv_of_processor_hartid (node, &hartid); +if (ret) +continue ; + +if (hartid <0 ) +continue ; + +/* set block-size for cbom extension if available */ + ret =of_property_read_u32 (node,"riscv,cbom-block-size" , &val); +if (ret) +continue ; + +if (!riscv_cbom_block_size) { + riscv_cbom_block_size = val; + cbom_hartid = hartid; + }else { +if (riscv_cbom_block_size != val) +pr_warn ("cbom-block-size mismatched between harts %d and %lu\n" , + cbom_hartid, hartid); + } + } +} +#endif arch/riscv/mm/dma-noncoherent.c +0
-31@@ -8,11 +8,8 @@#include <linux/dma-direct.h> #include <linux/dma-map-ops.h> #include <linux/mm.h> -#include <linux/of.h> -#include <linux/of_device.h> #include <asm/cacheflush.h> -static unsigned int riscv_cbom_block_size =L1_CACHE_BYTES ;static bool noncoherent_supported; void arch_sync_dma_for_device(phys_addr_t paddr, size_t size,@@ -75,41 +72,6 @@ void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size, dev->dma_coherent = coherent; } -#ifdef CONFIG_RISCV_ISA_ZICBOM -void riscv_init_cbom_blocksize (void ) -{ -struct device_node *node; -int ret; -u32 val; - -for_each_of_cpu_node (node) { -unsigned long hartid; -int cbom_hartid; - - ret =riscv_of_processor_hartid (node, &hartid); -if (ret) -continue ; - -if (hartid <0 ) -continue ; - -/* set block-size for cbom extension if available */ - ret =of_property_read_u32 (node,"riscv,cbom-block-size" , &val); -if (ret) -continue ; - -if (!riscv_cbom_block_size) { - riscv_cbom_block_size = val; - cbom_hartid = hartid; - }else { -if (riscv_cbom_block_size != val)arch/riscv/mm/dma-noncoherent.c void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size, · line 77
Quoted 10 lines
arch/riscv/mm/dma-noncoherent.c +0
-7-pr_warn ("cbom-block-size mismatched between harts %d and %lu\n" , - cbom_hartid, hartid); - } - } -} -#endif -void riscv_noncoherent_supported (void ) { noncoherent_supported = true;_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
PATCH v2 3/4 RISC-V: Implement arch specific PMEM APIs
In reply to Anup PatelThe NVDIMM PMEM driver expects arch specific APIs for cache maintenance and if arch does not provide these APIs then NVDIMM PMEM driver will always use MEMREMAP_WT to map persistent memory which in-turn maps as UC memory type defined by the RISC-V Svpbmt specification.
Now that the Svpbmt and Zicbom support is available in RISC-V kernel, we implement PMEM APIs using ALT_CMO_OP() macros so that the NVDIMM PMEM driver can use MEMREMAP_WB to map persistent memory.
--- arch/riscv/Kconfig | 1 + arch/riscv/mm/Makefile | 1 + arch/riscv/mm/pmem.c | 21 +++++++++++++++++++++ 3 files changed, 23 insertions(+) create mode 100644 arch/riscv/mm/pmem.c
arch/riscv/Kconfig +1 -0
@@ -25,6 +25,7 @@ config RISCV select ARCH_HAS_GIGANTIC_PAGE select ARCH_HAS_KCOV select ARCH_HAS_MMIOWB + select ARCH_HAS_PMEM_API select ARCH_HAS_PTE_SPECIAL select ARCH_HAS_SET_DIRECT_MAP if MMU select ARCH_HAS_SET_MEMORY if MMU
arch/riscv/mm/Makefile +1 -0
@@ -31,3 +31,4 @@ endif obj-$(CONFIG_DEBUG_VIRTUAL) += physaddr.o obj-$(CONFIG_RISCV_DMA_NONCOHERENT) += dma-noncoherent.o +obj-$(CONFIG_ARCH_HAS_PMEM_API) += pmem.o
arch/riscv/mm/pmem.c new file +21 -0
@@ -0,0 +1,21 @@ +// SPDX-License-Identifier: GPL-2.0 +/* +* Copyright (C) 2022 Ventana Micro Systems Inc. +*/ + +#include <linux/export.h> +#include <linux/libnvdimm.h> + +#include <asm/cacheflush.h> + +void arch_wb_cache_pmem (void *addr,size_t size) +{ +ALT_CMO_OP (clean, addr, size, riscv_cbom_block_size); +} +EXPORT_SYMBOL_GPL (arch_wb_cache_pmem); + +void arch_invalidate_pmem (void *addr,size_t size) +{ +ALT_CMO_OP (inval, addr, size, riscv_cbom_block_size); +} +EXPORT_SYMBOL_GPL (arch_invalidate_pmem);
Signature · 1 line
2.34.1
Re: [PATCH v2 3/4] RISC-V: Implement arch specific PMEM APIs
Hi Anup,
Am Dienstag, 30. August 2022, 06:46:41 CEST schrieb Anup Patel:
Quoted 8 lines
The NVDIMM PMEM driver expects arch specific APIs for cache maintenance and if arch does not provide these APIs then NVDIMM PMEM driver will always use
MEMREMAP_WTto map persistent memory which in-turn maps as UC memory type defined by the RISC-V Svpbmt specification.Now that the Svpbmt and Zicbom support is available in RISC-V kernel, we implement PMEM APIs using
ALT_CMO_OP()macros so that the NVDIMM PMEM driver can useMEMREMAP_WBto map persistent memory.
Zicbom is detected at runtime, though that kconfig setting changes the behaviour for the memremap-type at compile-time. So what happens on systems not using zicbom (or another cmo-variant) ?
Heiko
Quoted 59 lines
Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>--- arch/riscv/Kconfig | 1 + arch/riscv/mm/Makefile | 1 + arch/riscv/mm/pmem.c | 21 +++++++++++++++++++++ 3 files changed, 23 insertions(+) create mode 100644 arch/riscv/mm/pmem.carch/riscv/Kconfig +1
-0@@ -25,6 +25,7 @@ config RISCV select ARCH_HAS_GIGANTIC_PAGE select ARCH_HAS_KCOV select ARCH_HAS_MMIOWB + select ARCH_HAS_PMEM_API select ARCH_HAS_PTE_SPECIAL select ARCH_HAS_SET_DIRECT_MAP if MMU select ARCH_HAS_SET_MEMORY if MMUarch/riscv/mm/Makefile +1
-0@@ -31,3 +31,4 @@ endif obj-$(CONFIG_DEBUG_VIRTUAL) += physaddr.o obj-$(CONFIG_RISCV_DMA_NONCOHERENT) += dma-noncoherent.o +obj-$(CONFIG_ARCH_HAS_PMEM_API) += pmem.oarch/riscv/mm/pmem.c new file +21
-0@@ -0,0 +1,21 @@ +// SPDX-License-Identifier: GPL-2.0 +/* +* Copyright (C) 2022 Ventana Micro Systems Inc. +*/ + +#include <linux/export.h> +#include <linux/libnvdimm.h> + +#include <asm/cacheflush.h> + +void arch_wb_cache_pmem (void *addr,size_t size) +{ +ALT_CMO_OP (clean, addr, size, riscv_cbom_block_size); +} +EXPORT_SYMBOL_GPL (arch_wb_cache_pmem); + +void arch_invalidate_pmem (void *addr,size_t size) +{ +ALT_CMO_OP (inval, addr, size, riscv_cbom_block_size); +} +EXPORT_SYMBOL_GPL (arch_invalidate_pmem);
Re: [PATCH v2 3/4] RISC-V: Implement arch specific PMEM APIs
Quoted 9 lines · Heiko Stübner
Hi Anup,
Am Dienstag, 30. August 2022, 06:46:41 CEST schrieb Anup Patel:
Quoted 5 lines
The NVDIMM PMEM driver expects arch specific APIs for cache maintenance and if arch does not provide these APIs then NVDIMM PMEM driver will always use
MEMREMAP_WTto map persistent memory which in-turn maps as UC memory type defined by the RISC-V Svpbmt specification.
Quoted 7 lines
Quoted 3 lines
Now that the Svpbmt and Zicbom support is available in RISC-V kernel, we implement PMEM APIs using
ALT_CMO_OP()macros so that the NVDIMM PMEM driver can useMEMREMAP_WBto map persistent memory.Zicbom is detected at runtime, though that kconfig setting changes the behaviour for the memremap-type at compile-time. So what happens on systems not using zicbom (or another cmo-variant) ?
On a system without Zicbom (or some other cmo-variant), the PMEM read will always work but PMEM writes will not be reliable.
Currently, the generic PMEM driver has no provision to allow arch code to disable cacheable mapping at boot-time.
Maybe we can add WARN_ONCE() for the case when arch_xyz_pmem() is called on a system not having Zicbom ?
Regards,
Anup
Quoted 67 lines
Heiko
Quoted 59 lines
Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>--- arch/riscv/Kconfig | 1 + arch/riscv/mm/Makefile | 1 + arch/riscv/mm/pmem.c | 21 +++++++++++++++++++++ 3 files changed, 23 insertions(+) create mode 100644 arch/riscv/mm/pmem.carch/riscv/Kconfig +1
-0@@ -25,6 +25,7 @@ config RISCV select ARCH_HAS_GIGANTIC_PAGE select ARCH_HAS_KCOV select ARCH_HAS_MMIOWB + select ARCH_HAS_PMEM_API select ARCH_HAS_PTE_SPECIAL select ARCH_HAS_SET_DIRECT_MAP if MMU select ARCH_HAS_SET_MEMORY if MMUarch/riscv/mm/Makefile +1
-0@@ -31,3 +31,4 @@ endif obj-$(CONFIG_DEBUG_VIRTUAL) += physaddr.o obj-$(CONFIG_RISCV_DMA_NONCOHERENT) += dma-noncoherent.o +obj-$(CONFIG_ARCH_HAS_PMEM_API) += pmem.oarch/riscv/mm/pmem.c new file +21
-0@@ -0,0 +1,21 @@ +// SPDX-License-Identifier: GPL-2.0 +/* +* Copyright (C) 2022 Ventana Micro Systems Inc. +*/ + +#include <linux/export.h> +#include <linux/libnvdimm.h> + +#include <asm/cacheflush.h> + +void arch_wb_cache_pmem (void *addr,size_t size) +{ +ALT_CMO_OP (clean, addr, size, riscv_cbom_block_size); +} +EXPORT_SYMBOL_GPL (arch_wb_cache_pmem); + +void arch_invalidate_pmem (void *addr,size_t size) +{ +ALT_CMO_OP (inval, addr, size, riscv_cbom_block_size); +} +EXPORT_SYMBOL_GPL (arch_invalidate_pmem);
PATCH v2 4/4 RISC-V: Enable PMEM drivers
In reply to Anup PatelWe now have PMEM arch support available in RISC-V kernel so let us enable relevant drivers in defconfig.
--- arch/riscv/configs/defconfig | 1 + 1 file changed, 1 insertion(+)
arch/riscv/configs/defconfig +1 -0
@@ -159,6 +159,7 @@ CONFIG_VIRTIO_MMIO=y CONFIG_RPMSG_CHAR=y CONFIG_RPMSG_CTRL=y CONFIG_RPMSG_VIRTIO=y +CONFIG_LIBNVDIMM=y CONFIG_EXT4_FS=y CONFIG_EXT4_FS_POSIX_ACL=y CONFIG_EXT4_FS_SECURITY=y
Signature · 1 line
2.34.1
Re: [PATCH v2 4/4] RISC-V: Enable PMEM drivers
Quoted 8 lines · Anup Patel
EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
We now have PMEM arch support available in RISC-V kernel so let us enable relevant drivers in defconfig.
Co-developed-by Mayuresh Chitale <[email protected]>Signed-off-by Mayuresh Chitale <[email protected]>Signed-off-by Anup Patel <[email protected]>
Bunch of new build time warnings related to pmem w/ this patch but they all seem to be problems in the nvdimm core code that've just been exposed here...
Here's another name to add to your single-line defconfig change:
Quoted 19 lines
--- arch/riscv/configs/defconfig | 1 + 1 file changed, 1 insertion(+)arch/riscv/configs/defconfig +1
-1@@ -159,6 +159,7 @@ CONFIG_VIRTIO_MMIO=y CONFIG_RPMSG_CHAR=y CONFIG_RPMSG_CTRL=y CONFIG_RPMSG_VIRTIO=y +CONFIG_LIBNVDIMM=y CONFIG_EXT4_FS=y CONFIG_EXT4_FS_POSIX_ACL=y CONFIG_EXT4_FS_SECURITY=y --2.34.1